def __init__(self,
field_dims,
embed_dim,
normalize,
sparse_embedding,
mlp_dims,
method='inner'):
super().__init__()
num_fields = len(field_dims)
if method == 'inner':
self.pn = InnerProductNetwork()
elif method == 'outer':
raise NotImplementedError()
self.pn = OuterProductNetwork(num_fields, embed_dim)
else:
raise ValueError('unknown product type: ' + method)
self.embedding = FeaturesEmbedding(field_dims,
embed_dim,
sparse_embedding=sparse_embedding,
normalize=normalize)
self.linear = FeaturesLinear(field_dims,
embed_dim,
sparse_embedding=sparse_embedding)
self.embed_output_dim = num_fields * embed_dim
self.mlp = MultiLayerPerceptronV2(num_fields * (num_fields - 1) // 2 +
self.embed_output_dim,
mlp_dims,
normalize="none")
def __init__(self, field_dims, embed_dim, normalize, sparse_embedding, attn_size, dropouts):
super().__init__()
self.num_fields = len(field_dims)
self.embedding = FeaturesEmbedding(field_dims, embed_dim, normalize=normalize,
sparse_embedding=sparse_embedding)
self.linear = FeaturesLinear(field_dims, sparse_embedding=sparse_embedding)
self.afm = AttentionalFactorizationMachine(embed_dim, attn_size, dropouts)
def __init__(self, field_dims, embed_dim, normalize, sparse_embedding,
mlp_dims):
super().__init__()
self.linear = FeaturesLinear(field_dims,
sparse_embedding=sparse_embedding)
self.embedding = FeaturesEmbedding(field_dims,
embed_dim,
sparse_embedding=sparse_embedding,
normalize=normalize)
self.embed_output_dim = len(field_dims) * embed_dim
self.mlp = MultiLayerPerceptronV2(self.embed_output_dim, mlp_dims)
def __init__(self, field_dims, embed_dim, normalize, sparse_embedding,
LNN_dim, mlp_dims, dropouts):
super().__init__()
self.num_fields = len(field_dims)
self.linear = FeaturesLinear(
field_dims, sparse_embedding=sparse_embedding) # Linear
self.embedding = FeaturesEmbedding(
field_dims,
embed_dim,
normalize=normalize,
sparse_embedding=sparse_embedding) # Embedding
self.LNN_dim = LNN_dim
self.LNN_output_dim = self.LNN_dim * embed_dim
self.LNN = LNN(self.num_fields, embed_dim, LNN_dim)
self.mlp = MultiLayerPerceptron(self.LNN_output_dim, mlp_dims,
dropouts[0])
def __init__(self, field_dims, embed_dim, atten_embed_dim, num_heads, num_layers, mlp_dims, attention_dropout,
mlp_dropout, has_residual=True):
super().__init__()
self.num_fields = len(field_dims)
self.linear = FeaturesLinear(field_dims)
self.embedding = FeaturesEmbedding(field_dims, embed_dim)
self.atten_linear = torch.nn.Linear(embed_dim, atten_embed_dim)
self.embed_output_dim = len(field_dims) * embed_dim
self.atten_output_dim = len(field_dims) * atten_embed_dim
self.has_residual = has_residual
self.mlp = MultiLayerPerceptron(self.embed_output_dim, mlp_dims, mlp_dropout)
self.self_attns = torch.nn.ModuleList([
torch.nn.MultiheadAttention(atten_embed_dim, num_heads, dropout=attention_dropout) for _ in
range(num_layers)
])
self.attn_fc = torch.nn.Linear(self.atten_output_dim, 1)
if self.has_residual:
self.V_res_linear = torch.nn.Linear(embed_dim, atten_embed_dim)
def __init__(self,
field_dims,
embed_dim,
normalize,
sparse_embedding,
mlp_dims,
cross_layer_sizes,
split_half=True):
super().__init__()
self.embedding = FeaturesEmbedding(field_dims,
embed_dim,
normalize=normalize,
sparse_embedding=sparse_embedding)
self.linear = FeaturesLinear(field_dims,
sparse_embedding=sparse_embedding)
self.embed_output_dim = len(field_dims) * embed_dim
self.cin = CompressedInteractionNetwork(len(field_dims),
cross_layer_sizes, split_half)
self.mlp = MultiLayerPerceptronV2(self.embed_output_dim,
mlp_dims,
output_layer=True)
def __init__(
self,
field_dims,
embed_dim,
normalize,
sparse_embedding,
mlp_dims,
embedding_dropout=0.5,
hidden_dropout=0.3,
embedding_bn=True,
layer_bn=True,
):
super().__init__()
self.embedding = FeaturesEmbedding(field_dims,
embed_dim,
normalize=normalize,
sparse_embedding=sparse_embedding)
self.linear = FeaturesLinear(field_dims,
sparse_embedding=sparse_embedding)
if embedding_bn:
self.fm = torch.nn.Sequential(
FactorizationMachine(reduce_sum=False),
torch.nn.BatchNorm1d(embed_dim),
torch.nn.Dropout(embedding_dropout))
else:
self.fm = torch.nn.Sequential(
FactorizationMachine(reduce_sum=False),
torch.nn.Dropout(embedding_dropout))
self.mlp = MultiLayerPerceptronV2(
embed_dim,
mlp_dims,
output_layer=True,
dropout=hidden_dropout,
normalize="bn" if layer_bn else "none",
)